Saddle Points and Eigenvalue Escape Routes
At a saddle, the Hessian is a symmetric matrix whose eigenvalues are the second-derivative curvatures along its eigenvectors; positive eigenvalues mean stable directions and negative eigenvalues mean escape directions, with eigenvectors giving the exact axes of motion.
A complete interactive classroom, not just a preview.
Start when you are ready to enter this Stage's 7 scenes and explore, respond, and learn as you go.
How do the Hessian's eigenvalues reveal escape directions near a saddle?
At a saddle point the surface curves up in some directions and down in others, and the Hessian's eigenvalues tell you exactly which way is which.
A saddle looks flat if you only check the function value, so it feels stable — yet perturbations along certain directions make you slide downhill. How does the matrix of second derivatives encode that hidden instability?
A manipulable 3D saddle surface where the learner tilts the surface to change the eigenvalues and watches the eigenvector arrows align with the steepest escape and the unstable ridge.
Positive eigenvalues mark directions where the function curves upward (stable); negative eigenvalues mark directions where it curves downward (escape routes); the eigenvectors of the Hessian point exactly along those axes.
A common first guess is that the gradient tells you which way to escape, but the gradient is zero at the saddle itself — so the answer must live in the second derivatives encoded by the Hessian.
- Non-symmetric or degenerate Hessians
- Infinite-dimensional or functional saddle problems
- Optimization algorithms such as Newton's method or trust regions
- Connection to index theorems and Morse theory
- Numerical computation of Hessians for large-scale problems
- 01The Hidden Escape RouteslideQuestion
Frame the driving question with a visual of a saddle surface: along one ridge the surface falls away, along another it rises. Why does the Hessian know which direction is which?
- A saddle point is a critical point that is neither a minimum nor a maximum.
- The gradient vanishes there, so it cannot tell you which way is downhill.
- The answer must come from the second derivatives — the Hessian.
- 02Commit Your HypothesisquizPrediction
Ask the learner to choose what the Hessian's eigenvalues should look like at a saddle point before any explanation is revealed.
- Force a single committed guess.
- Surface the misconception that all eigenvalues share a sign.
- 03Reading the Hessian's Diagonal FormslideEvidence
Show the symmetric 2x2 Hessian of a simple saddle (e.g. f(x,y) = x^2 - y^2) and then its diagonalization, making the eigenvalues and eigenvectors explicit as a small data table.
- Hessian = [[2, 0], [0, -2]] in the natural basis.
- Eigenvalues are +2 and -2; eigenvectors are the x and y axes.
- Sign of eigenvalue matches sign of curvature along that eigenvector.
- 04Saddle Surface ExplorerinteractiveExplanation
Let the learner rotate the surface and switch between two saddles (z = x^2 - y^2 and z = 2x^2 - y^2) to see how eigenvectors rotate with the basis and how eigenvalue magnitudes control steepness.
- Drag to rotate the surface and confirm eigenvector alignment visually.
- Toggle between two saddle functions and observe the eigenvectors stay on the symmetry axes.
- See that eigenvector arrows lie exactly along the escape and stable directions.
- 05When the Hessian Stops TalkingslideBoundary
Show a rotated saddle f(x,y) = (x+y)^2 - (x-y)^2 to illustrate that the Hessian is basis-independent, and flag the degenerate case f(x,y) = x^2 where the Hessian is positive semidefinite and no escape direction exists.
- Rotating coordinates does not change the eigenvalues, only the eigenvectors.
- A zero eigenvalue means the Hessian is inconclusive along that direction.
- Non-symmetric or singular Hessians require extra care beyond this picture.
- 06Find the Escape Route on a Tilted SaddleinteractiveTransfer
Give the learner a saddle whose principal axes are rotated 30 degrees and ask them to drag an arrow to the escape direction; the widget checks whether the chosen direction matches the eigenvector of the negative eigenvalue.
- Apply the eigenvalue rule in a rotated, less obvious basis.
- Confirm that eigenvectors — not coordinate axes — are the true escape directions.
- Reveal the matching negative eigenvalue once the correct direction is chosen.
- 07The Hessian's VerdictslideResolution
Close the loop by directly answering the driving question: the Hessian's eigenvalues are the curvatures along its eigenvectors, negative eigenvalues mark escape directions, positive eigenvalues mark stable ones.
- Eigenvectors of the Hessian give the principal curvature axes.
- Sign of each eigenvalue = sign of curvature = stable or escape.
- Magnitude of each eigenvalue = how steep that escape or stability is.
- This is why second-order information is essential near saddle points.
Discussion threads for a Stage aren't available yet.
